{"id":"https://openalex.org/W4412945596","doi":"https://doi.org/10.18653/v1/2025.acl-long.444","title":"Improving Preference Extraction In LLMs By Identifying Latent Knowledge Through Classifying Probes","display_name":"Improving Preference Extraction In LLMs By Identifying Latent Knowledge Through Classifying Probes","publication_year":2025,"publication_date":"2025-01-01","ids":{"openalex":"https://openalex.org/W4412945596","doi":"https://doi.org/10.18653/v1/2025.acl-long.444"},"language":"en","primary_location":{"id":"doi:10.18653/v1/2025.acl-long.444","is_oa":false,"landing_page_url":"https://doi.org/10.18653/v1/2025.acl-long.444","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","raw_type":"proceedings-article"},"type":"article","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":null,"any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5025730671","display_name":"Sharan Maiya","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Sharan Maiya","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5014118666","display_name":"Yinhong Liu","orcid":"https://orcid.org/0009-0006-6723-0558"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yinhong Liu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5007263620","display_name":"Ramit Debnath","orcid":"https://orcid.org/0000-0003-0727-5683"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ramit Debnath","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5081393566","display_name":"Anna Korhonen","orcid":"https://orcid.org/0000-0002-3692-3144"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Anna Korhonen","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":4,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":{"value":0.08894667,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"9061","last_page":"9081"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10181","display_name":"Natural Language Processing Techniques","score":0.9922999739646912,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T10181","display_name":"Natural Language Processing Techniques","score":0.9922999739646912,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T11063","display_name":"Rough Sets and Fuzzy Logic","score":0.95169997215271,"subfield":{"id":"https://openalex.org/subfields/1703","display_name":"Computational Theory and Mathematics"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10215","display_name":"Semantic Web and Ontologies","score":0.9398999810218811,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/preference","display_name":"Preference","score":0.6349848508834839},{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.5443664193153381},{"id":"https://openalex.org/keywords/artificial-intelligence","display_name":"Artificial intelligence","score":0.3903151750564575},{"id":"https://openalex.org/keywords/data-science","display_name":"Data science","score":0.3259246051311493},{"id":"https://openalex.org/keywords/statistics","display_name":"Statistics","score":0.086994469165802},{"id":"https://openalex.org/keywords/mathematics","display_name":"Mathematics","score":0.06456324458122253}],"concepts":[{"id":"https://openalex.org/C2781249084","wikidata":"https://www.wikidata.org/wiki/Q908656","display_name":"Preference","level":2,"score":0.6349848508834839},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5443664193153381},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3903151750564575},{"id":"https://openalex.org/C2522767166","wikidata":"https://www.wikidata.org/wiki/Q2374463","display_name":"Data science","level":1,"score":0.3259246051311493},{"id":"https://openalex.org/C105795698","wikidata":"https://www.wikidata.org/wiki/Q12483","display_name":"Statistics","level":1,"score":0.086994469165802},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.06456324458122253}],"mesh":[],"locations_count":2,"locations":[{"id":"doi:10.18653/v1/2025.acl-long.444","is_oa":false,"landing_page_url":"https://doi.org/10.18653/v1/2025.acl-long.444","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","raw_type":"proceedings-article"},{"id":"pmh:doi:10.17863/cam.118290","is_oa":true,"landing_page_url":null,"pdf_url":null,"source":{"id":"https://openalex.org/S4406922384","display_name":"Open MIND","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Conference Object"}],"best_oa_location":{"id":"pmh:doi:10.17863/cam.118290","is_oa":true,"landing_page_url":null,"pdf_url":null,"source":{"id":"https://openalex.org/S4406922384","display_name":"Open MIND","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Conference Object"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W4391375266","https://openalex.org/W2899084033","https://openalex.org/W2748952813","https://openalex.org/W2390279801","https://openalex.org/W4391913857","https://openalex.org/W2358668433","https://openalex.org/W4396701345","https://openalex.org/W2376932109","https://openalex.org/W2001405890","https://openalex.org/W4396696052"],"abstract_inverted_index":{"Large":[0],"Language":[1],"Models":[2],"(LLMs)":[3],"are":[4],"often":[5],"used":[6],"as":[7],"automated":[8],"judges":[9],"to":[10,38],"evaluate":[11],"text,":[12],"but":[13],"their":[14],"effectiveness":[15],"can":[16,99],"be":[17,141],"hindered":[18],"by":[19,30],"various":[20],"unintentional":[21],"biases.":[22],"We":[23],"propose":[24],"using":[25,52],"linear":[26,113],"classifying":[27],"probes,":[28],"trained":[29],"leveraging":[31],"differences":[32],"between":[33],"contrasting":[34],"pairs":[35],"of":[36,54],"prompts,":[37],"directly":[39],"access":[40],"LLMs\u2019":[41],"latent":[42],"knowledge":[43],"and":[44,61,69,78,98,119,138],"extract":[45],"more":[46],"accurate":[47],"preferences.":[48],"Through":[49],"extensive":[50],"experiments":[51],"models":[53,132],"varying":[55],"size":[56],"from":[57],"four":[58],"different":[59],"families":[60],"six":[62],"diverse":[63],"datasets":[64],"assessing":[65],"text":[66],"quality":[67],"evaluation":[68],"common":[70],"sense":[71],"reasoning,":[72],"we":[73],"demonstrate":[74],"that":[75],"both":[76],"supervised":[77],"unsupervised":[79],"probing":[80,114],"approaches":[81],"consistently":[82],"outperform":[83,101],"traditional":[84],"generation-based":[85],"judgement":[86],"while":[87,126],"maintaining":[88],"similar":[89],"computational":[90],"costs.":[91],"These":[92],"probes":[93],"generalise":[94],"under":[95],"domain":[96],"shifts":[97],"even":[100],"finetuned":[102],"evaluators":[103],"with":[104],"the":[105,145],"same":[106],"training":[107],"data":[108,137],"size.":[109],"Our":[110,136],"results":[111],"suggest":[112],"offers":[115],"an":[116],"accurate,":[117],"robust":[118],"computationally":[120],"efficient":[121],"approach":[122],"for":[123],"LLM-as-judge":[124],"tasks":[125],"providing":[127],"interpretable":[128],"insights":[129],"into":[130],"how":[131],"encode":[133],"judgement-relevant":[134],"knowledge.":[135],"code":[139],"will":[140],"openly":[142],"released":[143],"in":[144],"future.":[146]},"counts_by_year":[],"updated_date":"2026-06-11T09:08:48.828518","created_date":"2025-10-10T00:00:00"}